| Age | Commit message (Collapse) | Author | |
|---|---|---|---|
| 2021-04-10 | Calculate pose similarity loss and canonical consistency loss of each part ↵ | Jordan Gong | |
| after pooling | |||
| 2021-04-06 | Turn off gradient when decoding images | Jordan Gong | |
| 2021-04-06 | Network modification | Jordan Gong | |
| This commit change the auto-encoder by removing fc and optimizing latent space features | |||
| 2021-03-22 | Add embedding visualization and validate on testing set | Jordan Gong | |
| 2021-03-14 | Bug fix when transforming and new config | Jordan Gong | |
| 2021-03-12 | Code refactoring | Jordan Gong | |
| 1. Separate FCs and triplet losses for HPM and PartNet 2. Remove FC-equivalent 1x1 conv layers in HPM 3. Support adjustable learning rate schedulers | |||
| 2021-02-27 | Implement Batch Hard triplet loss and soft margin | Jordan Gong | |
| 2021-02-26 | Fix predict function | Jordan Gong | |
| 2021-02-21 | Remove FConv blocks | Jordan Gong | |
| 2021-02-20 | Separate triplet loss from model | Jordan Gong | |
| 2021-02-18 | Implement adjustable input size and change some default configs | Jordan Gong | |
| 2021-02-18 | Decode mean appearance feature | Jordan Gong | |
| 2021-02-18 | Decode mean appearance feature | Jordan Gong | |
| 2021-02-15 | Revert "Memory usage improvement" | Jordan Gong | |
| This reverts commit be508061 | |||
| 2021-02-14 | Memory usage improvement | Jordan Gong | |
| This update separates input data to two batches, which reduces ~30% memory usage. | |||
| 2021-02-14 | Prepare for DataParallel | Jordan Gong | |
| 2021-02-09 | Improve performance when disentangling | Jordan Gong | |
| This is a HUGE performance optimization, up to 2x faster than before. Mainly because of the replacement of randomized for-loop with randomized tensor. | |||
| 2021-02-08 | Code refactoring, modifications and new features | Jordan Gong | |
| 1. Decode features outside of auto-encoder 2. Turn off HPM 1x1 conv by default 3. Change canonical feature map size from `feature_channels * 8 x 4 x 2` to `feature_channels * 2 x 16 x 8` 4. Use mean of canonical embeddings instead of mean of static features 5. Calculate static and dynamic loss separately 6. Calculate mean of parts in triplet loss instead of sum of parts 7. Add switch to log disentangled images 8. Change default configuration | |||
| 2021-01-23 | Remove the third term in canonical consistency loss | Jordan Gong | |
| 2021-01-23 | Transform all frames together in evaluation | Jordan Gong | |
| 2021-01-21 | Print average losses after 100 iters | Jordan Gong | |
| 2021-01-09 | Add prototype predict function | Jordan Gong | |
| 2021-01-09 | Change auto-encoder input in evaluation | Jordan Gong | |
| 2021-01-07 | Add typical training script and some bug fixes | Jordan Gong | |
| 1. Resolve deprecated scheduler stepping issue 2. Make losses in the same scale(replace mean with sum in separate triplet loss, enlarge pose similarity loss 10x) 3. Add ReLU when compute distance in triplet loss 4. Remove classes except Model from `models` package init | |||
| 2021-01-06 | Add TensorBoard support | Jordan Gong | |
| 2021-01-05 | Implement Batch All Triplet Loss | Jordan Gong | |
| 2021-01-05 | Change and improve weight initialization | Jordan Gong | |
| 1. Change initial weights for Conv layers 2. Find a way to init last fc in init_weights | |||
| 2021-01-03 | Separate last fc matrix from weight init function | Jordan Gong | |
| Recursive apply will override other parameters too | |||
| 2021-01-03 | Implement weight initialization | Jordan Gong | |
| 2021-01-03 | Update hyperparameter configuration, implement prototype fit function | Jordan Gong | |
| 2021-01-03 | Add separate fully connected layers | Jordan Gong | |
| 2021-01-02 | Separate training and evaluating | Jordan Gong | |
| 2021-01-02 | Correct feature dims after disentanglement and HPM backbone removal | Jordan Gong | |
| 1. Features used in HPM is decoded canonical embedding without transpose convolution 2. Decode pose embedding to image for Part Net 3. Backbone seems to be redundant, we can use feature map given by auto-decoder | |||
| 2021-01-02 | Change type of pose similarity loss to tensor | Jordan Gong | |
| 2020-12-31 | Implement some parts of RGB-GaitPart wrapper | Jordan Gong | |
| 1. Triplet loss function and weight init function haven't been implement yet 2. Tuplize features returned by auto-encoder for later unpack 3. Correct comment error in auto-encoder 4. Swap batch_size dim and time dim in HPM and PartNet in case of redundant transpose 5. Find backbone problems in HPM and disable it temporarily 6. Make feature structure by HPM consistent to that by PartNet 7. Fix average pooling dimension issue and incorrect view change in HP | |||
 |
index : gait-recognition.git | |
| Unnamed repository; edit this file 'description' to name the repository. | Git Version Control |
| summaryrefslogtreecommitdiff |